279 19 Visualization for Site Assessment Hiroyuki Kohsaka and Tomoko Sekine CONTENTS 19.1 Introduction 279 19.2 Multilevel Measures of Accessibility and Its Spatial Variation within Residential Districts 281 19.2.1 Accessibility Measured at the Residential-District Level 281 19.2.2 Accessibility Measured at 100 M Mesh Level 282 19.2.3 Visualization of Spatial Variation in Accessibility within a Residential District 284 19.2.3.1 Bivariate Map of Accessibility and Its Variability 285 19.2.3.2 Composite Map of Accessibility by Two-Level Visualization 287 19.2.3.3 Accessibility Map at Variable Spatial Level 289 19.3 Measure of Accessibility by Highly Accurate Simulation and Its Visualization 290 19.3.1 Population as Demand Volume 290 19.3.2 Development of Road Network 291 19.3.3 Measure of Navigation Road Distance by Highly Accurate Simulation Considering Complex Traffic Conditions 292 19.4 Conclusion 296 References 298 19.1 Introduction Numerous approaches for site assessment have been developed in geogra- phy to evaluate sites for housing and retail facilities (Orford, 1999; Jones and Simmons, 1990). These approaches evaluate a site in terms of two factors, 2713_C019.fm Page 279 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC 280 GIS-based Studies in the Humanities and Socail Sciences such as the site itself and its location. Site factor is related to the lot in which a facility may be located and the physical environment directly related to the facility. Location factor is connected with its surrounding, which provides opportunity of use or demand. Recently, site-assessment approaches have been performed on GIS to handle very complicated consumer markets (Bir- kin et al., 2004). Accessibility is one of the major elements for the location factor in site assessment. Accessibility is measured from two sides, demand and supply. The measure of accessibility from the residential site to retail and service facilities is related to evaluate a housing site from the demand side. Five types of accessibility measures have been proposed: 1) container index, 2) minimum distance, 3) cumulative opportunity, 4) gravity potential, and 5) space–time (Kwan 1998). Talen and Anselin (1998) point out that the choice of accessibility measure has to be considered very carefully using a case study of the geographic-accessibility measures to public playgrounds at the census-tract level. For site assessment to a retail or service facility, it is necessary to evaluate whether a site will be able to attract a certain volume of sales. Evaluation methods have been developed, such as 1) rating model, 2) regression model, and 3) spatial-interaction model (Birkin et al., 2002). The first is related to compare relative scores for sites, and the second and third can predict the sales volume using mathematical models. Accessibility for supply side is measured in site assessment for retail and service facilities. In the rating model, buffer technique is used to determine a straightforward, “accessible” area followed by overlay technique to clip out this buffer area. However this buffer/overlay approach has some shortcomings, in the point that transport network, natural or man-made barriers, and competition with already estab- lished outlets are not taken into account (Geertman, et al., 2004). However, this approach is widely used in practical site assessment by the reason of its simplicity (for example, site assessment for petrol forecourts is referred to in Birkin et al., 2003). When these site-assessment approaches are applied to practical scenes, many problems have been pointed out. One of the critical issues is the accuracy of analytical results. Inaccurate results cannot be guaranteed to clear the hurdle of a resident’s satisfaction or a client’s sales target. The reliability of the end result to reduce the risk of a wrong or misleading decision is important to site assessment (Van der Wel, et al., 1994). The decision-maker therefore wants to reveal the extent to which uncertainty affects the “decision space.” The presentation of uncertain information is one use of visualization in the GIS community. The extra visual attribute that a visualization environ- ment provides can be used to add a further dimension to a map, in order to judge “truth” on GIS by measures, of uncertainty, error (accuracy), variation, validity, reliability, stability, or probability (MacEachren, 1995). The visual- ization techniques to display uncertainty include side-by-side, overlay, and merged displays (Beard and Buttenfield, 1999). The merged display makes 2713_C019.fm Page 280 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC Visualization for Site Assessment 281 use of a bivariate map as a representation of quantitative data and reliability of those data. For example, visualization techniques are applied to convey classification of uncertainty in classified imagery and soil maps (Fisher, 1994a; 1994b). For classified imagery, the uncertainty inherent in the assignment of a pixel to a class is conveyed by making the value or color of a pixel proportionate to the strength of it belonging to a particular class. Gahegan (2000) depicts a false color satellite-image fragment of an agricultural area, where vertical offset is used to represent the probability (as determined by a classifier) of a pixel being classified as “wheat.” This paper tackles improving the accuracy of site assessment using suitable visualization techniques to reduce the risk of a misleading site selection. In the second section, the visualization is applied to display classification uncer- tainty in an accessibility map to ophthalmic clinics. The third section per- forms a highly accurate simulation as a site-assessment approach for a car dealer to reveal “truth” as an inaccessible site. The last section discusses a mechanism to judge whether a highly accurate approach should be applied in the practical scenes. 19.2 Multilevel Measures of Accessibility and Its Spatial Variation within Residential Districts 19.2.1 Accessibility Measured at the Residential-District Level As a case study in this section, accessibility is measured from residential districts (Cyocyo-aza) to ophthalmic clinics in Matsudo City, Chiba Prefec- ture, Japan. Matsudo is one of the satellite cities in the Tokyo metropolitan area. Its area is 61 square kilometers, and 19 ophthalmic clinics are located within the city. The shortest-path distance to the nearest clinic is measured using the second method in five accessibility measures mentioned above. Figure 19.1 shows location ( + ) of clinics and centroids of residential districts (residential point; ᭡ ) on the road network of the northwest part of Matsudo City. The shortest-path distance from each residential point to the nearest clinic is measured on the actual road network using network analysis of ArcView (Sekine, 2003). Figure 19.2 shows statistical distribution of the shortest-path distance for 343 residential districts. The average of the distance is 1177 m, and its stan- dard deviation is 575 m. By considering such a distribution of the distance, the degree of accessibility is divided into four accessibility levels, as follows: “good” is shorter than 750 m, “normal” is 750 m to 1500 m, “bad” is 1500 m to 2250 m, and “very bad” is longer than 2250 m. All residential districts in Matsudo City are classified into four levels of accessibility in Figure 19.3. 2713_C019.fm Page 281 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC 282 GIS-based Studies in the Humanities and Socail Sciences According to this result, residential district “A” shown in Figure 19.3 was assessed as “good” in terms of the accessibility to ophthalmic clinics. 19.2.2 Accessibility Measured at 100 M Mesh Level Now let us measure the accessibility at finer level. The 1-kilometer mesh constructed in the Basic Area Mesh System 1 is divided into 10 equal segments for each side to create 100 m mesh. The shortest-path distance to the clinics is measured from the centroids of 6089 100 m meshes constituting Matsudo City. Figure 19.4 shows the four accessibility levels at 100 m mesh level. The residential district A consists of 22 meshes, as shown in Figure 19.5. Fourteen meshes are assessed as “good” accessibility, and eight meshes are assessed as “normal.” Therefore, we can recognize variation in accessibility FIGURE 19.1 Location of ophthalmic clinics and centroids of residential districts on road network. 01 2 km Clinic Residential Point Road W N S E 2713_C019.fm Page 282 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC Visualization for Site Assessment 283 within this district. An issue may be raised by the residents in the meshes assessed as “normal,” because they will find that their residential place is “normal” in spite of being assessed as “good” at the district level. This is known as modifiable area-unit problem (MAUP) and is particularly impor- tant for the residents in the case of lowering the accessibility level. In this case, the site assessment gives wrong information to them. To examine such a variation in accessibility within all residential districts of Matsudo City, the accessibility map at the district level (Figure 19.3) was intersected with one at 100 m mesh level (Figure 19.4). Table 19.1 shows the variation volume of accessibility between two levels. The diagonal cells represent no change in accessibility level. These ratios are about 70 percent for “normal,” and about 60 percent for “good,” “bad,” and “very bad.” The ratios to the lower accessibility level are 33 percent for “good” and 15 percent for “normal.” Inversely, the ratios to raise the level are 13 percent for “normal,” 28 percent for “bad,” and 36 percent for “very bad.” For good or bad, it became clear that 30 percent to 40 percent of meshes have different accessibility levels from the one measured at the district level. The degradation of accessibility level, in other words, the rate at which spatial analysis at the district level over-assesses, amounts to 15 percent to 33 percent. And more than 30 percent of meshes within the district assessed as “very bad” raise the accessibility level. The result of this analysis means that the accessibility measured at the district level has not enough accuracy in practice. FIGURE 19.2 Shortest-path distance to the nearest ophthalmic clinics from residential points. Shortest path distance (m) 3500 3000 2500 2000 1500 1000 500 0 1 10 19 28 37 46 55 64 73 82 91 100 109 118 127 136 145 154 163 172 181 190 199 208 217 226 235 244 253 262 271 280 289 298 307 316 325 334 343 Rank 2713_C019.fm Page 283 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC 284 GIS-based Studies in the Humanities and Socail Sciences 19.2.3 Visualization of Spatial Variation in Accessibility within a Residential District Two stages of the selection process will be adopted in the selection of a residential site within a city. The first stage selects a residential district in the city, and the second selects a site within the district. Therefore, it is necessary to position the accessibility at an individual site or at the 100 m mesh level in the range of accessibility for the whole city, as shown at the residential-district level. To represent accessibility at the residential-district level while holding enough accuracy, three visualization techniques are pro- posed in the following. FIGURE 19.3 Accessibility to ophthalmic clinics at district level. A Clinic Accessibility 0 1 2 km N W E S ~750 m: Good 750 m~1,500 m: Normal 1,500 m~2,250 m: Bad 2,250 m~: Very bad 2713_C019.fm Page 284 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC Visualization for Site Assessment 285 19.2.3.1 Bivariate Map of Accessibility and Its Variability The first visualization is the overlay display in which accessibility and its variability is simultaneously represented as a bivariate map. Figure 19.6 is a bivariate map in which accessibility is classified into four levels, and its variability within the district is classified into four levels, such as 0 percent, 1 percent to 25 percent, 26 percent to 50 percent, and 51 percent or more. If the variability is zero, then accessibility is distributed uniformly within the district. If the variability is 51 percent or more, it means half or more of meshes consisting of the district differ from the accessibility level assessed at the district level. FIGURE 19.4 Accessibility to ophthalmic clinics at 100 m mesh level. Clinic Accessibility 0 1 2 km N S W E ~750 m: Good 750 m~1,500 m: Normal 1,500 m~2,250 m: Bad 2,250 m~ Very bad 2713_C019.fm Page 285 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC 286 GIS-based Studies in the Humanities and Socail Sciences This map shows that classification accuracy (uncertainty) is different even among the districts assessed as “good” accessibility (see district A, which is “good” in accessibility and is 26 percent to 50 percent in variability). There- fore, we can avoid making a misleading decision in evaluating a residential district using this map. However, this visualization shows the level of accu- FIGURE 19.5 Spatial variation in accessibility within residential district A. TABLE 19.1 Boundary Clinic Accessibility ~750 m: Good 750 m~1,500 m: Normal 1,500 m~2,250 m: Bad 2,250 m~: Very bad 0 N S W E 0.5 1 km A 100m mesh level District level Good Normal Bad Very bad Good 1521(67.6) 555(24.7) 125(5.5) 50(2.2) Normal 575(13.4) 3051(71.2) 630(14.7) 30(0.7) Bad 76(2.8) 688(24.9) 1683(61.0) 313(11.3) Very bad 0(0.0) 14(2.4) 195(33.4) 375(64.2) 2713_C019.fm Page 286 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC Visualization for Site Assessment 287 racy for the districts, but cannot show where and to what degree accessibility differs inside the district. 19.2.3.2 Composite Map of Accessibility by Two-Level Visualization The second is two-level visualization technique. Usually, a map is constructed at one level of spatial resolution. However, there may be a transitional zone in which the accessibility will be changed from one level to another level. For the district including such a zone, the result of accessibility should be repre- sented at more detail (high) spatial resolution. Two-level visualization will be used for such a situation to hold accuracy of the result. Now, let us apply two-level visualization to the accessibility in Matsudo City. If the accessibility level for a residential district is the same level for FIGURE 19.6 (See color insert following page 176.) Bivariate map of accessibility at district level and its internal variability. W S E N Accessibility/Variability 0 1 2 km Good/0% Good/1~25% Good/26~50% Good/51%~ Normal/0% Normal/1~25% Normal/26~50% Normal/51%~ Bad/0% Bad/1~25% Bad/26~50% Bad/51%~ Very bad/0% Very bad/1~25% Very bad/26~50% Very bad/51%~ A 2713_C019.fm Page 287 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC 288 GIS-based Studies in the Humanities and Socail Sciences 100m meshes consisting of it, the district is considered as a uniform area in terms of accessibility. In other words, no variability exists within the residential district. Then the analytical results at the district level are used. Contrarily, if a district includes 100 m meshes with different accessibility levels, the results at 100 m mesh level will be used, because spatial variation of the accessibility cannot be represented at the district level. Figure 19.7 shows the accessibility composed at two levels, depending on its spatial variability. In 261 districts 2 , the residential districts with homogeneous accessibility are 49 (18.8 percent). Namely, accessibility mea- sured at the district level has enough accuracy for about 20 percent of districts. The accessibility levels for their districts are broken down into 12 districts (24.5 percent) as “good,” 20 (40.8 percent) as “normal,” 15 (30.6 FIGURE 19.7 Composite map of accessibility at residential district level and 100 m mesh level. A W S N E Accessibility 0 1 2 km ~750 m: Good 750 m ~ 1,500 m: Normal 1,500 m ~ 2,250 m: Bad 2,250 m ~ : Very bad 2713_C019.fm Page 288 Monday, September 26, 2005 8:11 AM Copyright © 2006 Taylor & Francis Group, LLC [...]... Figure 19. 9 also shows the zones of 1 km and 2 km at sites A and B, defined by road distance using the “Find service area” tool in the network analysis of ArcView In comparison with the areas of 1 km and 2 km zones defined Copyright © 2006 Taylor & Francis Group, LLC 2713_C 019. fm Page 292 Monday, September 26, 2005 8:11 AM 292 GIS- based Studies in the Humanities and Socail Sciences TABLE 19. 2a TABLE 19. 2b... Group, LLC 2713_C 019. fm Page 294 Monday, September 26, 2005 8:11 AM 294 GIS- based Studies in the Humanities and Socail Sciences N 0 0.5 1 1.5 2 km W E S Navigation route Site A Signal Residential point Road 2 km Zone FIGURE 19. 10c Navigation route to site A in considering narrow road and one-way road shortest path by interacting between man and machine, namely, the “Shortest path finder” tool in ArcView Three... road Therefore, they make a detour, as shown in Figure 19. 10a The second obstacle factor is that the site faces on the outer (northward) side of Circle Road 8, which is one of industrial arterial roads in Tokyo The separation of the outer and inner sides by median strips makes site A inaccessible in turning right directly on Circle Road 8 Then the customer approaching from the inner side has to pass the. .. Demand Volume To measure demand volume at the first step of site assessment (Geertman et al., 2004), the buffer/overlay approach was applied to sites A and B to calculate the populations within circle zones of 1- and 2-kilometer radiuses by direct distance from sites A and B (Figure 19. 9) The populations at site A are about 32,000 in the 1 km zone and 140,000 in the 2 km zone (Table 19. 2a), and the. .. site and make a U-turn at the intersection 700 meters away, as shown in Figure 19. 10b The third factor is that there are many narrow roads and one-way roads in Setagaya Ward The road network, with many narrow and one-way roads, obstructs the passing of cars Figure 19. 10c shows an example of a detour for the neighboring resident to approach site A The navigation road distance from each residential point... (Table 19. 2a) The same populations at site B are 39,000 in the 1 km zone and 135,000 in the 2 km zone (Table 19. 2b) Therefore, the dull business at site A may be explained by a smaller demand than site B by 33,000 in the 2 km zone, in addition to taking into consideration less development of the road network 19. 3.3 Measure of Navigation Road Distance by Highly Accurate Simulation Considering Complex... dissolve processing on GIS Figure 19. 8 shows the distribution of accessibility by the visualization at the variable spatial level This efficiently represents the distribution of accessibility between and within the districts The districts with “good” access appear as pink, the districts with “normal” as yellow, the districts with “bad” as sky-blue, and the districts with “very bad” as blue The districts... New York, 199 9, p 219 233 Birkin, M., Clarke, G.P., and Clarke, M., Retail Geography and Intelligent Network Planning, Wiley, Chichester, 2002 Birkin, M., Boden, P., and Williams, J., Spatial decision support systems for petrol forecourts, in Planning Support Systems in Practice, Geertman, S and Stillwell, J., Eds., Springer, Berlin, 2003 Birkin, M., Clarke, G., Clarke, M., and Culf, R., Using spatial... considered in a highly accurate simulation of navigation routes to prove the inaccessible condition at site A The first obstacle factor is Tama River, consisting of the boundary between the metropolis of Tokyo and Kanagawa Prefecture Figure 19. 10a shows a navigation route for the resident in Kanagawa Prefecture They will cross Tama River by using “Futako” bridge instead of using “Daisan-keihin” as a toll... (Table 19. 2b) The population within 1 km and 2 km zones by navigation road distance are 7000 (22 percent of the population in the 1 km zone by direct distance) and 23,000 (16 percent) at site A, and 19, 000 (34 percentage) and 95,000 (about 60 percent) at site B, as shown in Tables 19. 2a and 19. 2b As the result of analysis by highly accurate simulation, it becomes clear that site A has 73,000 smaller demand . LLC 280 GIS- based Studies in the Humanities and Socail Sciences such as the site itself and its location. Site factor is related to the lot in which a facility may be located and the physical. Group, LLC 282 GIS- based Studies in the Humanities and Socail Sciences According to this result, residential district “A” shown in Figure 19. 3 was assessed as “good” in terms of the accessibility. strips makes site A inac- cessible in turning right directly on Circle Road 8. Then the customer approaching from the inner side has to pass the site and make a U-turn at the intersection 700 meters